'balance' is a Python package that is maintained and released by the Core Data Science Tel-Aviv team in Meta. 'balance' performs and evaluates bias reduction by weighting for a broad set of experimental and observational use cases.
Although balance is written in Python, you don't need a deep Python understanding to use it. In fact, you can just use this notebook, load your data, change some variables and re-run the notebook and produce your own weights!
This quickstart demonstrates re-weighting specific simulated data, but if you have a different usecase or want more comprehensive documentation, you can check out the comprehensive balance tutorial.
There are four main steps to analysis with balance:
Let's dive right in!
The following is a toy simulated dataset.
from balance import load_data
INFO (2023-05-24 12:56:51,562) [__init__/<module> (line 52)]: Using balance version 0.9.0
target_df, sample_df = load_data()
print("target_df: \n", target_df.head())
print("sample_df: \n", sample_df.head())
target_df:
id gender age_group income happiness
0 100000 Male 45+ 10.183951 61.706333
1 100001 Male 45+ 6.036858 79.123670
2 100002 Male 35-44 5.226629 44.206949
3 100003 NaN 45+ 5.752147 83.985716
4 100004 NaN 25-34 4.837484 49.339713
sample_df:
id gender age_group income happiness
0 0 Male 25-34 6.428659 26.043029
1 1 Female 18-24 9.940280 66.885485
2 2 Male 18-24 2.673623 37.091922
3 3 NaN 18-24 10.550308 49.394050
4 4 NaN 18-24 2.689994 72.304208
target_df.head().round(2).to_dict()
# sample_df.shape
{'id': {0: '100000', 1: '100001', 2: '100002', 3: '100003', 4: '100004'},
'gender': {0: 'Male', 1: 'Male', 2: 'Male', 3: nan, 4: nan},
'age_group': {0: '45+', 1: '45+', 2: '35-44', 3: '45+', 4: '25-34'},
'income': {0: 10.18, 1: 6.04, 2: 5.23, 3: 5.75, 4: 4.84},
'happiness': {0: 61.71, 1: 79.12, 2: 44.21, 3: 83.99, 4: 49.34}}
In practice, one can use pandas loading function(such as read_csv()) to import data into the DataFrame objects sample_df and target_df.
The first thing to do is to import the Sample class from balance. All of the data we're going to be working with, sample or population, will be stored in objects of the Sample class.
from balance import Sample
Using the Sample class, we can fill it with a "sample" we want to adjust, and also a "target" we want to adjust towards.
We turn the two input pandas DataFrame objects we created (or loaded) into a balance.Sample objects, by using the .from_frame()
sample = Sample.from_frame(sample_df, outcome_columns=["happiness"])
target = Sample.from_frame(target_df, outcome_columns=["happiness"])
WARNING (2023-05-24 12:56:51,768) [util/guess_id_column (line 111)]: Guessed id column name id for the data WARNING (2023-05-24 12:56:51,775) [sample_class/from_frame (line 259)]: No weights passed. Adding a 'weight' column and setting all values to 1 WARNING (2023-05-24 12:56:51,784) [util/guess_id_column (line 111)]: Guessed id column name id for the data WARNING (2023-05-24 12:56:51,798) [sample_class/from_frame (line 259)]: No weights passed. Adding a 'weight' column and setting all values to 1
If we use the .df property call, we can see the DataFrame stored in sample. We can see how we have a new weight column that was added (it will all have 1s) in the importing of the DataFrames into a balance.Sample object.
sample.df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 1000 entries, 0 to 999 Data columns (total 6 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 id 1000 non-null object 1 gender 912 non-null object 2 age_group 1000 non-null object 3 income 1000 non-null float64 4 happiness 1000 non-null float64 5 weight 1000 non-null int64 dtypes: float64(2), int64(1), object(3) memory usage: 47.0+ KB
We can get a quick overview text of each Sample object, but just calling it.
Let's take a look at what this produces:
sample
(balance.sample_class.Sample)
balance Sample object
1000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
target
(balance.sample_class.Sample)
balance Sample object
10000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
Next, we combine the sample object with the target object. This is what will allow us to adjust the sample to the target.
sample_with_target = sample.set_target(target)
Looking on sample_with_target now, it has the target atteched:
sample_with_target
(balance.sample_class.Sample)
balance Sample object with target set
1000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
target:
balance Sample object
10000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
3 common variables: gender,age_group,income
We can use .covars() and then followup with .mean() and .plot() (barplots and qqplots) to get some basic diagnostics on what we got.
We can see how:
print(sample_with_target.covars().mean().T)
source self target _is_na_gender[T.True] 0.088000 0.089800 age_group[T.25-34] 0.300000 0.297400 age_group[T.35-44] 0.156000 0.299200 age_group[T.45+] 0.053000 0.206300 gender[Female] 0.268000 0.455100 gender[Male] 0.644000 0.455100 gender[_NA] 0.088000 0.089800 income 6.297302 12.737608
print(sample_with_target.covars().asmd().T)
source self age_group[T.25-34] 0.005688 age_group[T.35-44] 0.312711 age_group[T.45+] 0.378828 gender[Female] 0.375699 gender[Male] 0.379314 gender[_NA] 0.006296 income 0.494217 mean(asmd) 0.326799
print(sample_with_target.covars().asmd(aggregate_by_main_covar = True).T)
source self age_group 0.232409 gender 0.253769 income 0.494217 mean(asmd) 0.326799
sample_with_target.covars().plot()
Next, we adjust the sample to the target. The default method to be used is 'ipw' (which uses inverse probability/propensity weights, after running logistic regression with lasso regularization).
# Using ipw to fit survey weights
adjusted_ipw = sample_with_target.adjust()
INFO (2023-05-24 12:56:52,662) [ipw/ipw (line 428)]: Starting ipw function INFO (2023-05-24 12:56:52,665) [adjustment/apply_transformations (line 257)]: Adding the variables: [] INFO (2023-05-24 12:56:52,666) [adjustment/apply_transformations (line 258)]: Transforming the variables: ['gender', 'age_group', 'income'] INFO (2023-05-24 12:56:52,676) [adjustment/apply_transformations (line 295)]: Final variables in output: ['gender', 'age_group', 'income'] INFO (2023-05-24 12:56:52,685) [ipw/ipw (line 462)]: Building model matrix INFO (2023-05-24 12:56:52,791) [ipw/ipw (line 486)]: The formula used to build the model matrix: ['income + gender + age_group + _is_na_gender'] INFO (2023-05-24 12:56:52,792) [ipw/ipw (line 489)]: The number of columns in the model matrix: 16 INFO (2023-05-24 12:56:52,792) [ipw/ipw (line 490)]: The number of rows in the model matrix: 11000 INFO (2023-05-24 12:56:52,800) [ipw/ipw (line 521)]: Fitting logistic model INFO (2023-05-24 12:56:54,489) [ipw/ipw (line 564)]: max_de: None INFO (2023-05-24 12:56:54,493) [ipw/ipw (line 594)]: Chosen lambda for cv: [0.0131066] INFO (2023-05-24 12:56:54,494) [ipw/ipw (line 602)]: Proportion null deviance explained [0.17168419]
adjusted_cbps = sample_with_target.adjust(method = "cbps")
INFO (2023-05-24 12:56:54,506) [cbps/cbps (line 409)]: Starting cbps function INFO (2023-05-24 12:56:54,509) [adjustment/apply_transformations (line 257)]: Adding the variables: [] INFO (2023-05-24 12:56:54,510) [adjustment/apply_transformations (line 258)]: Transforming the variables: ['gender', 'age_group', 'income'] INFO (2023-05-24 12:56:54,520) [adjustment/apply_transformations (line 295)]: Final variables in output: ['gender', 'age_group', 'income'] INFO (2023-05-24 12:56:54,630) [cbps/cbps (line 460)]: The formula used to build the model matrix: ['income + gender + age_group + _is_na_gender'] INFO (2023-05-24 12:56:54,632) [cbps/cbps (line 472)]: The number of columns in the model matrix: 16 INFO (2023-05-24 12:56:54,633) [cbps/cbps (line 473)]: The number of rows in the model matrix: 11000 INFO (2023-05-24 12:56:54,641) [cbps/cbps (line 535)]: Finding initial estimator for GMM optimization INFO (2023-05-24 12:56:54,722) [cbps/cbps (line 562)]: Finding initial estimator for GMM optimization that minimizes the balance loss INFO (2023-05-24 12:56:55,060) [cbps/cbps (line 597)]: Running GMM optimization WARNING (2023-05-24 12:56:55,633) [cbps/cbps (line 612)]: Convergence of gmm_loss function with gmm_init start point has failed due to 'Maximum number of function evaluations has been exceeded.' WARNING (2023-05-24 12:56:56,205) [cbps/cbps (line 630)]: Convergence of gmm_loss function with beta_balance start point has failed due to 'Maximum number of function evaluations has been exceeded.' INFO (2023-05-24 12:56:56,211) [cbps/cbps (line 726)]: Done cbps function
print(adjusted_ipw)
Adjusted balance Sample object with target set using ipw
1000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
target:
balance Sample object
10000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
3 common variables: gender,age_group,income
# the adjusted object will look the same as ipw
print(adjusted_cbps)
Adjusted balance Sample object with target set using cbps
1000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
target:
balance Sample object
10000 observations x 3 variables: gender,age_group,income
id_column: id, weight_column: weight,
outcome_columns: happiness
3 common variables: gender,age_group,income
We can get a basic summary of the results:
print(adjusted_ipw.summary())
Covar ASMD reduction: 59.7%, design effect: 1.897 Covar ASMD (7 variables): 0.327 -> 0.132 Model performance: Model proportion deviance explained: 0.172
print(adjusted_cbps.summary())
Covar ASMD reduction: 78.1%, design effect: 2.828 Covar ASMD (7 variables): 0.327 -> 0.072
We can see that CBPS did a better job in terms of ASMD reduction. Let's look at it per feature:
We see an improvement in the average ASMD. We can look at detailed list of ASMD values per variables using the following call.
print("ipw:")
print(adjusted_ipw.covars().asmd().T)
print("\ncbps:")
print(adjusted_cbps.covars().asmd().T)
ipw: source self unadjusted unadjusted - self age_group[T.25-34] 0.001085 0.005688 0.004602 age_group[T.35-44] 0.037455 0.312711 0.275256 age_group[T.45+] 0.129304 0.378828 0.249525 gender[Female] 0.133970 0.375699 0.241730 gender[Male] 0.109697 0.379314 0.269617 gender[_NA] 0.042278 0.006296 -0.035983 income 0.243762 0.494217 0.250455 mean(asmd) 0.131675 0.326799 0.195124 cbps: source self unadjusted unadjusted - self age_group[T.25-34] 0.055902 0.005688 -0.050214 age_group[T.35-44] 0.022217 0.312711 0.290494 age_group[T.45+] 0.092277 0.378828 0.286551 gender[Female] 0.042673 0.375699 0.333026 gender[Male] 0.068241 0.379314 0.311072 gender[_NA] 0.044535 0.006296 -0.038240 income 0.106142 0.494217 0.388075 mean(asmd) 0.071586 0.326799 0.255213
It's easier to learn about the biases by just running .covars().plot() on our adjusted object.
adjusted_ipw.covars().plot(library = "seaborn", dist_type = "kde")
adjusted_cbps.covars().plot(library = "seaborn", dist_type = "kde")
We can also use different plots, using the seaborn library, for example with the "kde" dist_type.
And get the design effect using:
print("ipw:")
print(adjusted_ipw.weights().design_effect())
print("\ncbps:")
print(adjusted_cbps.weights().design_effect())
ipw: 1.8973847221820574 cbps: 2.8277650359852715
print(adjusted_ipw.outcomes().summary())
adjusted_ipw.outcomes().plot()
1 outcomes: ['happiness']
Mean outcomes (with 95% confidence intervals):
source self target unadjusted self_ci target_ci unadjusted_ci
happiness 53.389 56.278 48.559 (52.183, 54.595) (55.961, 56.595) (47.669, 49.449)
Response rates (relative to number of respondents in sample):
happiness
n 1000.0
% 100.0
Response rates (relative to notnull rows in the target):
happiness
n 1000.0
% 10.0
Response rates (in the target):
happiness
n 10000.0
% 100.0
The estimated mean happiness according to our sample is 48 without any adjustment and 54 with adjustment. The following show the distribution of happinnes:
print(adjusted_cbps.outcomes().summary())
adjusted_cbps.outcomes().plot()
1 outcomes: ['happiness']
Mean outcomes (with 95% confidence intervals):
source self target unadjusted self_ci target_ci unadjusted_ci
happiness 54.364 56.278 48.559 (52.983, 55.746) (55.961, 56.595) (47.669, 49.449)
Response rates (relative to number of respondents in sample):
happiness
n 1000.0
% 100.0
Response rates (relative to notnull rows in the target):
happiness
n 1000.0
% 10.0
Response rates (in the target):
happiness
n 10000.0
% 100.0
As we can see, CBPS has a larger design effect, but also fixes more of the ASMD and has an impact on the outcome. So there are pros and cons for each of the two methods.
Finally, we can prepare the data to be downloaded for future analyses.
adjusted_cbps.to_download()
# We can prepare the data to be exported as csv - showing the first 500 charaacters for simplicity:
adjusted_cbps.to_csv()[0:500]
'id,gender,age_group,income,happiness,weight\n0,Male,25-34,6.428659499046228,26.043028759747298,6.2503230109624734\n1,Female,18-24,9.940280228116047,66.88548460632677,0.4807350882449963\n2,Male,18-24,2.6736231547518043,37.091921916683006,2.223548307600131\n3,,18-24,10.550307519418066,49.39405003271002,4.454120253787142\n4,,18-24,2.689993854299385,72.30420755038209,3.203676675764448\n5,,35-44,5.995497722733131,57.28281646341816,16.803197782204208\n6,,18-24,12.63469573898972,31.663293445944596,5.423252165'
# Sessions info
import session_info
session_info.show(html=False, dependencies=True)
----- balance 0.9.0 pandas 1.4.3 session_info 1.0.0 ----- PIL 9.5.0 apport_python_hook NA argcomplete NA asttokens NA attr 21.2.0 backcall 0.2.0 beta_ufunc NA binom_ufunc NA cffi 1.15.1 colorama 0.4.4 comm 0.1.3 coxnet NA cvcompute NA cvelnet NA cvfishnet NA cvglmnet NA cvglmnetCoef NA cvglmnetPredict NA cvlognet NA cvmrelnet NA cvmultnet NA cycler 0.10.0 cython_runtime NA dateutil 2.8.2 debugpy 1.6.7 decorator 5.1.1 defusedxml 0.7.1 elnet NA executing 1.2.0 fastjsonschema NA fishnet NA glmnet NA glmnetCoef NA glmnetControl NA glmnetPredict NA glmnetSet NA glmnet_python NA hypergeom_ufunc NA idna 3.3 ipfn NA ipykernel 6.23.1 ipython_genutils 0.2.0 jedi 0.18.2 joblib 1.2.0 jsonpointer 2.0 jsonschema 3.2.0 kiwisolver 1.4.4 loadGlmLib NA lognet NA matplotlib 3.7.1 matplotlib_inline 0.1.6 mpl_toolkits NA mrelnet NA nbformat 5.8.0 nbinom_ufunc NA ncf_ufunc NA netifaces 0.11.0 numpy 1.24.3 packaging 23.1 parso 0.8.3 patsy 0.5.3 pexpect 4.8.0 pickleshare 0.7.5 pkg_resources NA platformdirs 3.5.1 plotly 5.14.1 prompt_toolkit 3.0.38 psutil 5.9.5 ptyprocess 0.7.0 pure_eval 0.2.2 pvectorc NA pydev_ipython NA pydevconsole NA pydevd 2.9.5 pydevd_file_utils NA pydevd_plugins NA pydevd_tracing NA pygments 2.15.1 pyparsing 2.4.7 pyrsistent NA pytz 2022.1 rfc3339_validator 0.1.4 rfc3986_validator 0.1.1 scipy 1.9.1 seaborn 0.11.1 sitecustomize NA six 1.16.0 sklearn 1.2.2 sphinxcontrib NA stack_data 0.6.2 statsmodels 0.14.0 tenacity NA threadpoolctl 3.1.0 tornado 6.3.2 traitlets 5.9.0 typing_extensions NA wcwidth 0.2.6 wtmean NA zmq 25.0.2 zoneinfo NA zope NA ----- IPython 8.13.2 jupyter_client 8.2.0 jupyter_core 5.3.0 notebook 6.5.4 ----- Python 3.10.6 (main, Mar 10 2023, 10:55:28) [GCC 11.3.0] Linux-5.15.0-1037-azure-x86_64-with-glibc2.35 ----- Session information updated at 2023-05-24 12:56